xf86-video-intel 2.5 release planning

Carl Worth cworth at cworth.org
Fri Aug 1 14:44:21 PDT 2008


On Fri, 2008-08-01 at 07:23 -0700, Eric Anholt wrote:
> On Thu, 2008-07-31 at 21:30 +0100, Colin Guthrie wrote:
> > For me, even on the 2.4 series, EXA is still unusably slow on my i945 
> > and compositing. Perhaps I'm missing a setting. It's certainly better 
> > than before (it used to crash the server!), but I'm still a little 
> > concerned about removing XAA completely (main issue right now is 
> > scrolling, but I'll need to read Carl's blog and see what's what!)

So, for reference, the stuff on my blog post shouldn't affect scrolling
on 945 systems.

> EXA is great on my 945-class systems, including "scrolling."  We really
> need sysprof output, and if pixmap migration is at the top of your
> sysprof output, then the list of what operations are failing to be
> accelerated from turning on the fallback debug flags in *_render.c and
> exa_priv.h.

Arjan just gave me a good test case that I was able to replicate on a
945 system here. The test case is:

	Install xfce4
	Go to "Right click"->Settings->Window manager tweaks->Compositor
	Set:
        * Enable display compositing
        * Show shadows under regular windows
     	* Opacity of windows during move (put at about halfway)

Then drag around a terminal window.

It starts out fairly snappy and then starts lagging horribly, (which
seems consistent with pixmap migration failures). Meanwhile, the same
operation is quite snappy with XAA. I'll follow up with the sysprof and
debugging output requested.

Note that "x11perf -compwinwin500" doesn't capture this EXA/XAA
performance difference, (it shows EXA nearly twice as fast as XAA on the
same machine).

-Carl





More information about the xorg mailing list